Your log file shows these messages appearing in a regular basis:
librdkafka: Global error: AllBrokersDown (Local: All broker connections are down): 1/1 brokers are down
Ignored event 'Error' on base producer poll
Currently, these messages appear in the log file if the data flow hasn't received any data from its input for four or five minutes. This does result in a request to the Event Hubs endpoint to keep the connection alive.
Can you double check that when you're using MQTT Explorer to manually publish a message that you're publishing to the correct topic and that the data source in your data flow is configured to subscribe to this topic?